The Trw Type IV Secretion System of Bartonella Mediates Host-Specific Adhesion to Erythrocytes

Bacterial pathogens typically infect only a limited range of hosts; however, the genetic mechanisms governing host-specificity are poorly understood. The α-proteobacterial genus Bartonella comprises 21 species that cause host-specific intraerythrocytic bacteremia as hallmark of infection in their respective mammalian reservoirs, including the human-specific pathogens Bartonella quintana and Bartonella bacilliformis that cause trench fever and Oroya fever, respectively. Here, we have identified bacterial factors that mediate host-specific erythrocyte colonization in the mammalian reservoirs. Using mouse-specific Bartonella birtlesii, human-specific Bartonella quintana, cat-specific Bartonella henselae and rat-specific Bartonella tribocorum, we established in vitro adhesion and invasion assays with isolated erythrocytes that fully reproduce the host-specificity of erythrocyte infection as observed in vivo. By signature-tagged mutagenesis of B. birtlesii and mutant selection in a mouse infection model we identified mutants impaired in establishing intraerythrocytic bacteremia. Among 45 abacteremic mutants, five failed to adhere to and invade mouse erythrocytes in vitro. The corresponding genes encode components of the type IV secretion system (T4SS) Trw, demonstrating that this virulence factor laterally acquired by the Bartonella lineage is directly involved in adherence to erythrocytes. Strikingly, ectopic expression of Trw of rat-specific B. tribocorum in cat-specific B. henselae or human-specific B. quintana expanded their host range for erythrocyte infection to rat, demonstrating that Trw mediates host-specific erythrocyte infection. A molecular evolutionary analysis of the trw locus further indicated that the variable, surface-located TrwL and TrwJ might represent the T4SS components that determine host-specificity of erythrocyte parasitism. In conclusion, we show that the laterally acquired Trw T4SS diversified in the Bartonella lineage to facilitate host-restricted adhesion to erythrocytes in a wide range of mammals.


Introduction
The successful infection of a mammalian host by a bacterial pathogen typically involves a series of intimate host-pathogen interactions. On the molecular level this is reflected by specific receptor-ligand interactions between bacterial virulence factors and their targeted host factors [1]. Adaptation of a bacterial virulence factor to a host factor that displays variability within the host population can restrict the host range that is susceptible to infection. The resulting host-specificity is an inherent feature of most bacterial pathogens of humans, including Helicobacter pylori, Listeria monocytogenes, Neisseria gonorrhoae, Salmonella typhi, Streptococcus pyogenes and Staphylococcus aureus. However, remarkably little is known about the molecular determinants of host specificity in bacterial infections, with the only exception of L. monocytogenes for which the conjugated action of two distinct host-specific invasion proteins was shown to be critical for fetoplacental listeriosis [2,3,4].
Bartonellae represent an interesting but largely unexplored model for host specificity. These facultative intracellular bacteria use arthropod transmission and hemotropism as mammalian parasitism strategies [5]. As the result of an adaptive radiation each of the 21 species infects only one or a few closely related mammalian reservoir host(s), which is highlighted by their capacity to cause a long-lasting intraerythrocytic bacteremia [6]. Nonreservoir hosts may get incidentally infected without developing an intraerythrocytic infection [7]. Two Bartonella species are humanspecific: Bartonella bacilliformis causes the biphasic Carrion's disease, with acute Oroya fever followed by the chronic verruga peruana, and Bartonella quintana causing trench fever. The life-threatening Oroya fever and the much milder course of trench fever represent the characteristic intraerythrocytic stages of these pathogens. The other 19 species cause intraerythrocytic infections in various nonprimate mammalian reservoirs. At least seven of them are recognized as zoonotic pathogens which incidentally infect humans. Commonly, B. henselae is associated with cat scratch disease [7].
The life cycle of Bartonella in the reservoir host has been analyzed in detail in rats experimentally infected with B. tribocorum [8]. Following intravenous inoculation, bacteria initially infect a primary niche outside of circulating blood, which is considered to comprise the vascular endothelium and possibly other cell types. Approximately on day five of infection, large numbers of bacteria are released into the bloodstream where they bind to and invade mature erythrocytes. Bacteria then replicate in a membranebound compartment until reaching a critical number. For the remaining life span of the erythrocytes the intracellular bacteria remain in a non-dividing state [8]. Monitoring of bacteremia in other animal models, such as the B. birtlesii-mouse [9] and B. henselae-cat models [10], or in captive naturally infected animals has yielded results that match those observed in the B. tribocorumrat model, suggesting a common mode of infection of the different species in their respective animal reservoirs [11]. The only exception is B. bacilliformis, which causes lysis of the infected human erythrocytes, eventually resulting in a severe hemolytic anemia.
The B. tribocorum-rat model was further explored to identify bacterial pathogenicity factors that are required for colonization of the mammalian reservoir host. A signature-tagged mutagenesis (STM) screen identified 98 essential bacterial loci [12], including genes encoding components of two distinct type IV secretion systems (T4SS), VirB/VirD4 and Trw, the invasion-associated locus B (IalB) protein, the trimeric autotransporter adhesin BadA, as well as further members of the autotransporter family [6]. Whether any of the identified genes is critical for host-specificity is unknown, although it is conceivable to assume that host-specificity loci are essential for infection and may thus be represented among the hits of the performed STM screen.
Experimental infections of different mammalian hosts by a given Bartonella strain have reproduced the species-specificity of erythrocyte invasion as observed in natural infections [11,13,14,15]. However, despite their availability, in vitro erythrocyte infection assays [16,17] have not been investigated for the study of host specificity. Here, we demonstrate for the first time that host specificity is reflected by the exclusive capacity of Bartonella species to adhere to erythrocytes isolated from their natural host(s). Second, by performing STM in Bartonella birtlesii followed by screening in mice in vivo and in isolated erythrocytes in vitro we identified the T4SS Trw as the molecular determinant of host-specific erythrocyte infection.

Results
An in vitro erythrocyte colonization assay to study host-restricted infection Based on described in vitro models of human and feline erythrocyte infection by B. bacilliformis and B. henselae, respectively [16,17], we established for B. birtlesii an in vitro infection model for erythrocytes isolated from the murine reservoir host. Balb/C mice were used as the source of erythrocytes as they are known to develop a long lasting intraerythrocytic infection upon experimental infection with B. birtlesii [9]. The intraerythrocytic presence of bacteria was evaluated over a period of three days using the gentamicin protection assay (Fig. 1A). Bacterial entry into erythrocytes was dependent on the number of bacteria per erythrocyte (multiplicity of infection, MOI; tested MOI range: 0.01 to 10) and time of infection (days post infection, DPI; tested time range: 1 to 3 DPI). The highest intraerythrocytic bacterial content over time was obtained for MOI = 0.1 and 1, with approximately 2610 5 colony forming units (CFU) per 10 10 erythrocytes (<0.002% infected erythrocytes) at 3 DPI. Given that mouse blood contains approximately 10 10 erythrocytes/ml, this value corresponds well to the bacteremia reported for experimentally infected Balb/C mice (<1610 5 CFU/ml; 0.001% infected erythrocytes) [9]. For MOI = 10, erythrocytes were infected at 1 DPI, but lysed entirely by 3 DPI. At MOI = 0.01, only low numbers of intraerythrocytic bacteria were detected over time. Based on these data, MOI = 1 was used for all subsequent erythrocyte infection assays. To evaluate whether the increase of intraerythrocytic bacteria over time was mainly due to continued bacterial invasion, or to intraerythrocytic bacterial multiplication, or to a combination of both, erythrocytes were infected with B. birtlesii for one day in the absence of gentamicin, followed by incubation in the continuous presence of gentamicin to kill extracellular bacteria. Fig. 1B shows that the number of intracellular bacteria increased over time in the presence of gentamicin, albeit to a lesser extent than in the untreated control. Bacteria thus appear to enter erythrocytes beyond 1 DPI and, moreover, to replicate in an intra-erythrocytic location.
Invasion of erythrocytes by Bartonella is preceded by bacterial adhesion to the erythrocyte surface [18]. To quantify erythrocytes infected by adherent extracellular and/or intracellular bacteria, we used GFP-expressing bacteria in combination with flow cytometry (Fig. 1C, D.). Similar as described for intraerythrocytic bacteria in the gentamicin protection assay, erythrocyte colonization revealed by flow cytometry was dependent on time ( Fig. 1.C) and MOI ( Fig. 1.D). However, the rate of erythrocyte colonization evaluated by flow cytometry (55% for MOI = 1 at 3 DPI) was approximately 209000-fold higher than erythrocytes invasion determined by the gentamicin protection assay (compare Fig. 1.A), indicating that the vast majority of bacteria detected by flow cytometry were associated extracellularly with erythro-

Author Summary
Pathogens are-as the result of adaptive evolution in their principal host(s)-typically limited in the range of hosts that they can infect successfully. However, infrequently such host-restricted pathogens may undergo a spontaneous host switch, which can lead to the evolution of pathogens with altered host specificity. Most human pathogens evolved this way, and animal-specific pathogens have thus to be considered as an important reservoir for the emergence of novel human pathogens. Despite host-specificity representing a common feature of pathogens, the underlying molecular mechanisms are largely unknown. In this study we have used bacterial pathogens of the genus Bartonella to identify bacterial factors involved in the determination of host specificity. The bartonellae represent an excellent model to study hostspecificity as each species is adapted to cause an intracellular infection of erythrocytes exclusively in its respective reservoir host(s). Using a genetic approach in combination with erythrocyte infection models in vitro and in vivo we demonstrate that a surface-located bacterial nanomachine-a so-called type IV secretion systemdetermines host specificity of erythrocyte infection. Our work sheds light on the molecular basis of host specificity and establishes an experimental model for studying the evolutionary processes facilitating spontaneous host shifts.
Next we investigated whether Bartonella species differ in their capacity to interact in vitro with erythrocytes of different mammalian origin, and whether this capacity may reflect the host-restriction displayed during natural infection. First, mouse erythrocytes were infected with either B. birtlesii, B. vinsonii arupensis (both mouse-specific), B. alsatica (rabbit-specific), B. vinsonii berkhoffii (dog-specific), B. henselae (cat-specific), B. quintana (human-specific), or B. tribocorum (rat-specific). Erythrocyte invasion was quantified by the gentamicin protection assay (Fig. 2). B. vinsonii arupensis displayed invasion rates similar to B. birtlesii, while none of the other strains tested resulted in significant erythrocyte invasion. Using a corresponding set of strains expressing GFP, consistent results were obtained for the flow cytometric determination of bacterial adhesion to mouse erythrocytes ( Fig. 3 and Fig. S1). These findings indicate that specificity for the mouse reservoir in vivo correlates with efficient adhesion to and invasion of mouse erythrocytes in vitro.
Next, we tested whether -similarly as observed for mouse erythrocytes and B. birtlesii -the capacity of B. henselae, B. quintana and B. tribocorum to adhere to erythrocytes in vitro is also restricted to erythrocytes from their natural reservoir host, i.e. cat, human and rat, respectively. GFP-expressing bacteria were used for erythrocyte infection, and adhesion was quantified by flow cytometric analysis. Fig. 3 and Table 1 illustrate that all tested Bartonella species were able to efficiently adhere to erythrocytes isolated from their respective reservoir hosts, while they essentially did not adhere to erythrocytes from non-reservoir hosts. The only exception is B. quintana, which further to erythrocytes from the human reservoir also colonized cat erythrocytes. Together, these data indicate that the established in vitro model of erythrocyte colonization reflects well the host restriction as observed during natural infection.

B. birtlesii genes required for intra-erythrocytic bacteremia in mice
As a basis for identifying genetic factors involved in hostrestricted erythrocyte colonization, we identified a comprehensive set of B. birtlesii genes required for establishing intraerythrocytic bacteremia in mice. To this end, an STM library of B. birtlesii was constructed as previously described for B. tribocorum [6,12,19]. From each conjugation assay, we selected 96 single kanamycinresistant colonies and assembled an STM mutant library of 3456 mutants. We then identified mutants that have lost the capacity to cause intraerythrocytic bacteremia by screening the library in the mouse infection model [9]. Of 1456 mutants tested in the input pools, 98 were not detected in the output pools from mice at days 7 and 14 post infection and were thus classified as abacteremic mutant candidates. All 98 abacteremic mutant candidates were retested by reassembling them into 49 pools of 9 mutants, each pool containing two abacteremic mutant candidates and an invariable set of seven mutants displaying wild-type behavior (bacteremic mutants). The rescreen confirmed an abacteremic phenotype for 48 of the initial 98 abacteremic mutant candidates, corresponding to 3.3% of the total number of mutants screened. Growth of all of the 48 confirmed abacteremic mutants on solid media was similar to the parental wild-type strain (data not shown).
We determined the transposon insertion sites for all 48 abacteremic mutants by direct sequencing out of the transposon into the flanking chromosomal region and mapping of the derived sequences onto the draft genome sequence of B. birtlesii (S. Cescau, H.M. Yang, J. Wang, M. Vayssier-Taussat, A. Danchin, and F. Biville, unpublished data). Three mutants harboring two separate transposon insertions were not considered further in the analysis. Table S1 lists the loci inactivated by single transposon insertion in the remaining 45 abacteremic mutants. Five mutants carried the transposon insertion in an intergenic region: one (83D04) was near a gene encoding a tRNA; three of them (04A01, 86C05, 69B07) were upstream of genes encoding proteins of unknown function and one (69C09) was in proximity to a putative transcriptional regulator gene. In these mutants, the transposon may have thus disrupted a promoter or another regulatory sequence. 40 transposon insertions were mapped to the coding region of 38 different protein-encoding genes. In 8 mutants the insertions were found in genes encoding a conserved protein of unknown function, among them three putative surface proteins. Sixteen mutants carried insertions in genes previously implicated in bacterial  pathogenicity, either in Bartonella (virB/D4, trw, ialA/B, badA, omp43, iba) or other pathogenic bacteria (i.e. loci encoding heat shock proteins) [5]. Moreover, mutant genes encoding proteins involved in transport and metabolism, as well as phage-related function were also identified.

B. birtlesii genes required for erythrocytic infection in vitro
The 45 confirmed abacteremic mutants with single transposon insertion were individually tested for their capacity to invade murine erythrocytes using the gentamicin protection assay (Table  S1). Nine mutants were found to be impaired in murine erythrocyte invasion (Fig. 4). Complementary erythrocyte adhesion assays based on flow cytometric analysis of antibody-stained bacteria demonstrated that seven of these nine invasion-deficient mutants are also deficient in erythrocytes adhesion. All seven mutants harbor a mutation in the operon encoding the T4SS Trw (two in trwD, trwE, trwF, trwJ2, trwL1, trwL2), which was previously shown to be important for establishing an intraerythrocytic bacteremia in B. tribocorum [20]. Compared to wild-type, both trwD mutants (04B03 and 41C12) showed a five-fold decrease in invasion/adhesion efficiency. All other trw mutants failed to invade erythrocytes and were severely impaired in their capacity to adhere to erythrocytes (Fig. 4). These data demonstrate that the Trw system is required for erythrocyte invasion by mediating specific adhesion to erythrocytes. In contrast, mutants harboring an insertion in the invasion-associated locus ialA/B showed normal erythrocyte adhesion but impaired invasion (10-fold reduced, p,0.01), confirming the previously suggested role of this locus in erythrocyte invasion [21,22]. Equally, an insertion mutant (25A02) inactivating livG (encoding an amino acid ABC-transporter) showed normal adhesion but a specific defect in invasion (4-fold, p,0.05) compared to wild-type (Fig. 4). None of the other abacteremic mutants appeared to be involved in erythrocytes invasion indicating that they probably are required for an earlier step of infection, i.e. for colonization of the primary niche.  [15], c [41,42], d [8,43,44], e [8].   Figure S2) or at 1 DPI by the gentamicin protection assay, respectively. The efficiency of erythrocyte adhesion or invasion of each tested mutant is expressed as percentage of erythrocyte adhesion or invasion of the isogenic wild-type strain (mean +/-SD of triplicate samples). All mutants listed in Role of Trw T4SS in host-specific infection of erythrocytes Next we tested whether the Trw system shown here to be essential for erythrocyte infection in vitro and in vivo may be directly involved in determining host-specificity. To this end we introduced pAB2, a plasmid encoding the trw locus of rat-specific B. tribocorum [20], into cat-specific B. henselae, human-specific B. quintana and B. tribocorum (control). We then compared the capacity of these recombinant strains to infect rat erythrocytes with their parental strains. Among the parental strains, only B. tribocorum mediated invasion of rat erythrocytes (Fig. 5), which is consistent with the erythrocyte adhesion data presented in Fig. 2. The pAB2mediated ectopic over-expression of trw in B. tribocorum resulted only in a slight increase of invasion, indicating that the endogenous level of trw expression is sufficient to mediate efficient bacterial entry. Strikingly, ectopic expression of the B. tribocorum trw locus in B. henselae and B. quintana rendered these pathogens capable of infecting rat erythrocytes. These data clearly demonstrate a direct role of the Trw system in determining host-specificity of erythrocyte infection.
To further assess which components of the Trw T4SS may mediate host specificity we analyzed the molecular evolution of different trw genes of B. birtlesii and related species. Candidate genes for mediating host specificity are surface exposed components, i.e. the T4SS pilus components TrwL and TrwJ. As shown by Nystedt et al. [23] for other Bartonella species, trwL and trwJ genes have been amplified and diversified several times during evolution. The trw locus of B. birtlesii also displays amplification of trwL (five copies) and co-amplification of trwJ together with trwH and trwI (two copies) (Fig. S3, panel A). Phylogenetic analyses and calculation of the non-synonymous (dN) and synonymous (dS) substitution frequencies of different trw genes further showed that trwJ and trwL homologs have diversified to much higher degree than other components of the Trw T4SS, within and among different species [Fig. S3, panel B-G, and [23]].

Discussion
Host-specificity is a prominent feature of pathogenic bacteria that reflects the host range susceptible to infection. Subtle changes in the molecular mechanisms that govern host-specificity may result in spontaneous host shifts, which represent a major risk for the emergence of novel human pathogens from animal reservoirs. Striking examples for this evolutionary scenario are the bartonellae, which cause host-restricted intra-erythrocytic infections in their mammalian reservoirs. In conjunction with repeated host shifts, the large number of Bartonella species evolved by adaptive radiation [6], including the human-specific pathogen B. quintana that evolved from cat-specific B. henselae [24]. Here we explored the bacterial genetic basis for host-restricted infection of erythrocytes. The establishment of an in vitro model of erythrocyte adherence and invasion allowed us to demonstrate for the first time a direct correlation of host-restricted erythrocyte infection in vivo and in vitro, demonstrating that host-specificity is determined by the capacity of bacteria to adhere to erythrocytes. In order to identify the bacterial factors critical for host-restricted erythrocyte infection we have used a two-step experimental protocol. First, we performed an STM screen for B. birtlesii in mice which allowed us to identify 45 abacteremic mutants defective in establishing intraerythrocytic infection. Among the corresponding set of 38 proteinencoding genes, 13 loci were also indentified in a similar STM screen performed in the B. tribocorum-rat model [6]. This indicates extensive similarities in the repertoire of pathogenesis factors in these closely related organisms as well as robustness of the performed genetic screens. Second, rescreening of the entire set of 45 abacteremic B. birtlesii mutants in the in vitro mouse erythrocyte infection model resulted in the identification of nine mutants impaired in erythrocte invasion. The other mutants (36 of 45 = 80%) displaying a wild-type phenotype in this assay are therefore not directly involved in erythrocyte infection, but rather may contribute to the establishment of infection in the primary niche. Prominent examples are the virB/virD4 genes encoding the VirB/VirD4 T4SS, which is known to be required for primary niche infection in the B. tribocorum/rat model (Schulein, 2002). Moreover, a recent study inferred the VirB/VirD4 T4SS as major bacterial factor facilitating bacterial adaptation to novel hosts [6]. The nine mutants impaired in in vitro erythrocyte invasion differ in their capacity to adhere to erythrocytes. Transposon insertions in the invasion locus (ialA/B) previously implicated in erythrocyte invasion [21,22] and livG encoding an amino acid ABCtransporter displayed wild-type like adherence to erythrocytes. IalA/B and LivG should thus represent invasion factors. The remaining seven invasion-deficient transposon mutants were all severely impaired in erythrocyte adhesion. Strikingly, all these mutants carry insertions in components encoding the T4SS Trw, which thus represents an erythrocyte adherence system that is critical for erythrocyte invasion. Trw is known to be required for establishing intra-erythrocytic infection in the B. tribocorum-rat model [5,20,25], however, evidence for a direct role of the Trw system in erythrocyte adhesion as provided here was lacking so far. Based on the presumable surface location of components of Trw [20] this T4SS may directly interact with the erythrocyte surface and thus may restrict the host range of erythrocyte infection. To test the hypothesis that Trw determines host range we have expressed Trw of rat-specific B. tribocorum in cat-specific B. henselae and human-specific B. quintana. Strikingly, this genetic manipulation resulted in an extension of the host range for in vitro erythrocyte infection towards rats, demonstrating that Trw indeed represents a major determinant of host-specificity of erythrocyte infection. Thus, this finding establishes a new experimental model to study the molecular mechanisms governing host restrictionfurther to the molecular paradigm of host-specificity exemplified by the interaction of two surface proteins of L. monocytogenes, InlA and InlB, with their respective host receptors [2,3,4].
The Trw locus was laterally acquired during evolution of the bartonellae. It is present in the largest sub-branch of the genus tree, comprising 13 species that are adapted to diverse mammalian reservoir hosts, while it is absent from humanspecific B. bacilliformis, cat-specific Bartonella clarridgeiae and the species of the ruminent-specific sub-branch, which all diverted early during evolution of the bartonellae [6]. Interestingly, the acquisition of Trw by the modern lineage correlates with the loss of flagella, which are know to represent a major pathogencity factor for the invasion of erythroctes by B. bacilliformis and probably other flagellated bartonellae [25]. The Trw system of Bartonella represents an interesting example of a pathogenesisrelated T4SS that evolved rather recently by functional diversification of a laterally acquired bacterial conjugation system. The trw locus displays characteristic features of a pathogenicity island and shares extensive similarity with the trw locus of IncW broad-host range plasmid R388 encoding a genuine conjugation system. The trw loci of Bartonella and R388 are colinear, except for multiple tandem gene duplications of trwL and trwJ-trwH in Bartonella. Complementation of R388 derivatives carrying mutations in different trw genes with their Bartonella homologues allowed to demonstrate functional interchangability for some T4SS components [20,26], underscoring the structural and functional conservation of individual subunits of these functionally diversified T4SSs. However, a major difference between these homologous systems is the lack of the coupling protein TrwB in Bartonella, which in R388 is required for export of T4SS substrates. The lack of TrwB in Bartonella thus indicates that its Trw system may not be capable of translocating substrates. However, the multiple copies of trwL and trwJ in the Bartonella trw locus encode variant forms of surface-exposed pilus components, which probably are all co-expressed [20], indicating that the primary function of the Bartonella Trw system may be the formation of variant pilus forms [25]. Based on the essential role of the Trw system for adhesion to erythrocyte and its role in determining host range it is conceivable to assume that these variant pili may facilitate the specific interaction with polymorphic erythrocyte receptors, either within the reservoir host population (e.g. different blood group antigens), or among different reservoir hosts. Phylogenetic analyses and calculation of the non-synonymous (dN) and synonymous (dS) substitution frequencies of different trw genes indeed demonstrated that trwJ and trwL homologs have diversified to much higher degree than other components of the Trw T4SS, within and among different species [23]. Together with the notion that the number of tandem repeats of trwL and trwJIH are variable among different Bartonella species these findings indicate that trwL and trwJ genes have been amplified and diversified several times during evolution. Horizontal transfer of such genes from a different bartonellae -similarly as we have demonstrated here for the entire trw operon of rat-specific B. tribocorum resulting in an extension of the host range of cat-specific B. henselae or humanspecific B. quintana to rat -or alternatively pre-adaption of superfluous copies of trwL and trwJ may represent realistic molecular evolutionary scenarios for host shifts and thereby the evolution of pathogens with an altered host-specificity as it has happened repeatedly during the evolution of the bartonellae. Future studies should identify the nature of the erythrocyte receptors targeted by the Trw system and their specific interaction that facilitate host-specific erythrocyte infection.

Ethics statement
Animals were handled in strict accordance with good animal practice as defined by the relevant European (European standards of welfare for animals in research), national (Information and  guidelines for animal experiments and alternative [30] were grown for 5 days on Columbia agar containing 5% defibrinated sheep blood (CBA) in a humidified atmosphere with 5% C0 2 at 35uC.

Construction of bacterial strains
B. tribocorum-gfp containing a chromosomally-integrated gfpexpression cassette [8] was used as GFP-expressing B. tribocorum strain. GFP-expressing bacteria of other Bartonella species were obtained by electroporation with plasmid pJMBGFP as previously described [31,32]. This plasmid was extracted and purified from B. quintana using a Midi Prep Kit (Qiagen). The electroporation procedures was described previously [31]. Transformed bacteria were selected by plating on CBA-Km. A signature-tagged mutant library of B. birtlesii IBS135 T was constructed as described for B. tribocorum [6,12]. Cosmid pAB2 encoding the entire trw locus of B. tribocorum [20,33,34] was introduced into B. henselae and B. quintana by three parental mating [34,35].

In vitro infection of erythrocytes
Erythrocytes from peripheral blood of mice (Balb/C), cats, rats (Wistar) and humans were isolated and purified by Ficoll gradient centrifugation. After washing in PBS, they were maintained in F12 modified medium [supplemented with 10% fetal calf serum, 2 mM glutamine, 1 mM sodium pyruvate, 0.1 mM Hepes, 257 mM histidine, 0.1 mg/ml hematin/histidine, non-essential amino acid (Gibco, FRANCE)] at 2610 8 /ml. For in vitro infection experiments, Bartonella species were grown on CBA or CBA-km (Bartonella-gfp and STM mutants) plates. After 5 days of culture (10 days for GFPexpressing Bartonella), bacteria were harvested, washed, suspended in PBS, and added to erythrocytes at a multiplicity of infection (MOI, calculation based on 1 OD 600 nm = 3610 9 bacteria/ml) varying from 0.01 to 10 and incubated at 35uC in 5% C0 2 for various periods of time (from 1 to 3 days).

Detection of erythrocyte-associated bacteria
Colonization of erythrocytes by Bartonella was assessed and quantified by the gentamicin protection assay, flow cytometry and confocal microscopy. For the quantification of intracellular bacteria by gentamicin protection, 100 ml were withdrawn from the invasion mixtures after 1, 2 or 3 days of in vitro infection. Mouse erythrocytes were separated from non-associated bacteria by washing 3 times with PBS and centrifuged at 500 g for 5 min. Erythrocytes were then incubated for 2 h at 35uC with gentamicin sulfate (250 mg/ml) to kill residual extracellular bacteria. Erythro-cytes were then washed three times in PBS to remove the antibiotic and intracellular bacteria were released from erythrocytes by hypotonic lyses of erythrocytes in 10 ml of sterile water by freezing at 220uC for 15 min. After thawing, serial dilutions of bacteria in PBS were inoculated onto CBA plates and incubated at 35uC for 5 days before being counted. For data presentation, all measurements were expressed as the number of CFU/10 10 erythrocytes (corresponding to <1 ml of blood).
For flow cytometric detection of erythrocyte-associated with GFP-expressing bacteria, measurements were performed at day 1, 2, 3 after in vitro infection of erythrocytes with ten days old bacterial cultures. 100 ml of the infection mixtures was washed 3 times in PBS and fixed for 10 min with 0.8% paraformaldehyde and 0.025% glutaraldehyde. After fixing, erythrocytes were analyzed by flow cytometry (FACScan, Becton Dickinson Bioscience, France). For flow cytometric detection of erythrocytes associated with bacteria that do not express GFP (abacteremic mutants), measurements were performed at day 2 after in vitro infection. 100 ml of the infection mixtures was washed 3 times in PBS and fixed for 10 min with 0.8% paraformaldehyde and 0.025% glutaraldehyde. After fixing, erythroctes/mutants association was revealed with mouse anti-B. birtlesii serum and antimouse FITC antibodies (Santa Cruz Biotechnology, Santa Cruz, CA,USA) and analyzed by flow cytomtry. Data were analyzed using the CellQuestPro software, version 4.0.2. Data for 109000 gated erythrocytes were collected and analyzed.
For confocal microscopy, 100 ml of the infection mixtures was washed three times in PBS and the erythrocytes cell surface was stained using goat anti-mouse GPA antibodies (Santa Cruz Biotechnology, California, USA) and labelled with anti-goat-PEantibodies (ImmunoQuest Antibody, North Yorkshire, UK). Samples were viewed with a Nikon Eclipse C1 Plus confocal laser scanning microscope (Nikon, Amstelveen, Netherlands) with detection in channel 1 (GFP fluorescence) and channel 2 (PE fluorescence) at original magnification x100.

STM library
The transposon vectors pHS006-Tag-001 to pHS006-Tag-036 each contained an oriT for conjugative transfer, the Himar1 transposon, a kanamycin resistant marker, a hyperactive transposase and one of 36 distinct signature-tags [6]. These 36 signature-tagged mariner transposon vectors were separately transferred from E. coli b2155 to B. birtlesii by two-parental mating as previously described [35]. From each mating, 96 single kanamycin-resistant B. birtlesii transconjugants were transferred to a 96-well plate with cryo-medium and stored at 280uC.

Mouse infections
Eight weeks old female Balb/C mice from Charles River Laboratories were housed in an animal facility (2 animals/cage) and allowed to acclimate to the facility and the diet for at least 5 days prior infection. Food and water were provided ad libitum. 36 differently signature-tagged mutants were grown separately from the transposon library for each input pool. They were pooled in PBS immediately before infection, and used to infect two mice with a total inoculum of 5610 7 colony forming units (10 ml of OD 595 = 1) in the ear dermis of Balb/C mice. The remainder of the input pools was heated at 100uC for 10 min and used as template for PCR detection. Fifty ml of blood were taken from the tail vein of the infected mice when bacteremia is peaking (days 7 and 14 postinfection) [9]. Bacteria released from erythrocytes by a freeze/thaw cycle were plated on CBA-km. After 10 days, bacterial colonies (output pool) were counted, harvested in PBS, suspended to OD 595 = 1 and heated at 100uC for 10 min to be used as template for PCR detection. The rescreen was done following the same protocol using pools of nine mutants (two abacteremic mutants and seven mutants displaying a wild-type phenotype).

PCR detection of abacteremic mutants
For signature-tag identification, the generic primer Srev01 corresponding to a sequence in the transposon and a set of tagspecific primer were used for amplification of a fragment of approximately 600 bp [6]. The conditions for the PCR were as follows: a first denaturation step at 95uC for 5 min, followed by 30 cycles of PCR with denaturation at 95uC for 1 min, annealing for 30 s at 52uC, and extension at 72uC for 1 min. The program was completed by an extension step at 72uC for 5 min. The amplified fragments were displayed on a 1% agarose gel. Mutants that were detected in the input pools and absent from the out put pools (days 7 and 14) in both mice were considered as abacteremic mutants.

Identification and analysis of transposon insertion sites
Genomic DNA from abacteremic mutants, regrown from the library, was prepared with the ROCHE Genomic DNA Isolation Kit. Genomic DNA was sent to QIAGEN for sequencing with primers Tn start and Tn end [6]. The sequences obtained by the genomic sequencing were compared by BlastN to the nr data base of NCBI (http://blast.ncbi.nlm.nih.gov/Blast.cgi). The exact transposon insertion sites were found by comparing the genomic sequences to contigs of the ongoing B. birtlesii genome sequencing project by BlastN.

Screening of abacteremic mutants for their capacity to infect murine erythrocytes
Mutants displaying an abacteremic phenotype were tested for their capacity to invade murine erythrocytes using the gentamicin protection assay. Each mutant was tested at MOI = 1 in at least two independent experiments performed in triplicate samples. For mutants displaying an impaired erythrocyte invasion phenotype, invasion assays were performed at least three times in triplicate samples and adhesion assays were tested at day 2 post infection by flow cytometric detection once in triplicate samples.

Statistical analysis
Numerical data are reported as the mean of at least 3 replicate samples +/standard errors of the means. Statistical significance of the data was measured by use of Student's t test. A p-value ,0.05 was considered significant.

Phylogenetic and evolutionary analysis
The sequence of the B. birtlesii trw locus was deposited under the EMBL-EBI accession no. FN555106. Sequence alignments were calculated with ClustalW as implemented in MEGA4. Phylogenetic trees were inferred by maximum likelihood methods with Paup 4.0 [36] and 100 bootstrap replicates were calculated. To select an appropriate substitution model the Akaike information criterion of Modeltest 3.7 was used [37]. The models obtained were general time reversible (GTR) + I for trwFED and trwN, transversion model (TVM) + I for trwI, and TVM + I + G for trwJ and trwL. Nonsynonymous (dN) and synonymous (dS) substitution frequencies were calculated using the method of Yang and Nielson [38] as implemented in the PAML package [39,40]. Figure S1 Efficiency of in vitro adhesion of murine erythrocytes to Bartonella sp. Freshly isolated murine erythrocytes were infected with Bartonella sp.-GFP (MOI = 1, detection at two DPI). The percentage of erythrocytes associated with bacteria were quantified by flow cytometric analysis. Representative data for the fluorescence (FL-1) of 10'000 erythrocytes are shown as histogram plots. Found at: doi:10.1371/journal.ppat.1000946.s001 (0.08 MB TIF) Figure S2 Efficiency of in vitro adhesion of murine erythrocytes to abacteremic mutants. Freshly isolated murine erythrocytes were infected with B. birtlesii abacteremic mutants (MOI = 1, detection 2 DPI). Association between erythrocytes and bacteria was revealed with mouse anti-B. birtlesii polyclonal serum and labelled with antimouse FITC antibody. The percentage of erythrocyte associated with bacteria was quantified by flow cytometric analysis. Representative data for the fluorescence (FL-1) of 10'000 erythrocytes are shown as histogram plots. according to their gene order. For the tandem repeated genes trwL, trwJ, trwI, and trwH only trwL5, trwJ1, trwI1, and trwH1 are shown, since ortholog assignment is difficult for the others due to copy number variation and the occurrence of recombination among different species [23]. Found at: doi:10.1371/journal.ppat.1000946.s003 (0.46 MB JPG)

Supporting Information
Table S1 Genotypic characterization of abacteremic mutants of B. birtlesii obtained by signature-tagged mutagenesis (STM). The columns BARBAKC, BH, BQ and BT list the extensions of systematic names of orthologous genes from the published genomes of B. bacilliformis (accession no. CP000524), B. henselae (accession no. BX897699), B. quintana (accession no. 897700) and B. tribocorum (accession no. AM260525), respectively. * The in vitro erythrocate invasion phenotype of each mutant was determined by the gentamicin protection assay after 1 day of infection (triplicate samples) and categorized as normal (.70% of wild-type), reduced (,70% of wild-type but .1% of wild-type) or none (,1% of wildtype). Mutants with reduced or none in vitro invasion were tested again (n = 3) and the resulting mean and SD of all three experiments are represented in Figure 4.